English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Delineating infarcted tissue in ischemic stroke lesions is crucial todetermine the extend of damage and optimal treatment for this life-threateningcondition. However, this problem remains challenging due to high variability ofischemic strokes' location and shape. Recently, fully-convolutional neuralnetworks (CNN), in particular those based on U-Net, have led to improvedperformances for this task. In this work, we propose a novel architecture thatimproves standard U-Net based methods in three important ways. First, insteadof combining the available image modalities at the input, each of them isprocessed in a different path to better exploit their unique information.Moreover, the network is densely-connected (i.e., each layer is connected toall following layers), both within each path and across different paths,similar to HyperDenseNet. This gives our model the freedom to learn the scaleat which modalities should be processed and combined. Finally, inspired by theInception architecture, we improve standard U-Net modules by extendinginception modules with two convolutional blocks with dilated convolutions ofdifferent scale. This helps handling the variability in lesion sizes. We splitthe 93 stroke datasets into training and validation sets containing 83 and 9examples respectively. Our network was trained on a NVidia TITAN XP GPU with 16GBs RAM, using ADAM as optimizer and a learning rate of 1$\times$10$^{-5}$during 200 epochs. Training took around 5 hours and segmentation of a wholevolume took between 0.2 and 2 seconds, as average. The performance on the testset obtained by our method is compared to several baselines, to demonstrate theeffectiveness of our architecture, and to a state-of-art architecture thatemploys factorized dilated convolutions, i.e., ERFNet.

Dense Multi-path U-Net for Ischemic Stroke Lesion Segmentation in Multiple Image Modalities